Forum Discussion

🚨 This forum is archived and read-only. To submit a forum post, please visit our new Developer Forum. 🚨
joan's avatar
joan
Protege
11 years ago

Looking for omnistereo benchmark CG scenes

I'm looking for benchmark scenes to evaluate/explore various algorithms and approaches to render stereoscopic spherical scenes to images (stereo cubemaps or equirectangular pairs).

The idea would be to have semi-simplistic scenes available in common formats or for all major renderers/technologies for comparison purposes.

The scenes should be specifically crafted to test the common artifacts that these capture and reprojections may yield: stitching seams, discontinuities towards the poles, wrong parallax, missing or faulty screen-space effects, etc.

Do such benchmark scenes already exist ?

I did create test scenes myself but I think it would be interesting to have a rigorous approach to it rather than putting a bunch of solids on a plane, and to have publicly available files so that anyone can perform the tests on their favorite renderer or plug-in.

Various strategies can be explored depending on whether the engine is a path tracer or a real time engine, if we need screen-space effects, and the chosen approach to pole merging…

Here are some examples of renders by various vendors/techs to illustrate the need for comparability (some link to large files).
- Demo stereo cubemaps in OTOY's OctaneVR.
- Demo equirect side by side video in Arnold renderer.
- Demo stereo cubemap in VRay.
- Demo equirect o/u in Blender Cycles test build.
- Test scene equirect o/u in unknown.
- Demo equirect o/u by OliVR script in Unity.
- Demo equirect o/u by eVRydayVR's "360 Panorama Capture" in Unity.
- Demo equirect in Unreal.

I wish we could compare them all on a common ground and also use it to experiment new approaches. Some kind of acid test for omnistereo rendering.

3 Replies

  • I haven't found a full sphere yet, but yes, it should be possible to merge two hemispheres from dome renderers.

    Thanks for the link it is instructive in several respects. 1, He describes such a test scene crafted to review artifacts ; 2, The claims of being better than other systems are a case in point, with a standard scene it would be easy to verify.
    It's not clear when the paper was written, I think the pole merging approach where the IPD is progressively reduced by a function of latitude would yield a similar radial pattern.